As with any DR product, there is no substitute for testing it. It depends how aggressive you want to be, but I'm...
into simply pulling the plug on a server and seeing if it fails over. Simulate the disaster as well as you can, and see how long it takes to get the application up and running in the cloud.
The best case is to take more of a high-availability (HA) approach where you have multiple physical hosts that can serve that workload. There are few companies that are experimenting with cloud and physical-based HA schemes, but I would say it is more likely that you'd simply build an HA application in the cloud and have the primary and the secondary copies living there.
Most clouds are designed with a scale-out methodology, so if any server fails or any component fails, it picks up where it left off on a different machine.
As companies get more familiar with HA and cloud-based HA schemes, I think we'll see some of the traditional DR methodologies go away. I think we'll see more of a cluster-based HA scheme.
How you should conduct a DR test
Why DR tests are important
Traditional DR test models have outgrown their usefulness
Dig Deeper on Disaster recovery services - outsourcing
Related Q&A from Andrew Reichman
With storage developments moving from hardware toward software, many vendors are reworking products to capitalize on the software-defined technology ... Continue Reading
When it comes to provisioning storage for virtual machines, techniques such as golden images and thin provisioning are key. Continue Reading
In physical environments data protection often means restoring an entire system, but using the same methodology for virtual servers is inefficient. Continue Reading